An isometric wireframe mesh of connected nodes with two highlighted traffic routes

The orchestration layer, drawn honestly: routes fail, queues absorb, nodes carry on.

Engineering note

Agent infrastructure that holds up in imperfect conditions

6 September 20263 min read

Most agent platforms assume the cloud is a given. Reliable bandwidth, stable power, data that can leave the building whenever it wants. In Namibia, none of those assumptions survive contact with a real site. Our agent infrastructure research starts somewhere else: what does it take for autonomous software to run on hardware you own, inside walls you control, when the link to the outside world is slow or gone?

Why self-hosted

The Tangison Agent platform, our in-house research platform formerly developed under the name Nangula AI, was built on one position: an AI workforce should run inside the client's infrastructure, not ours. Data stays where it is. Every action the agent takes is logged where the client can audit it.

The platform was discontinued as a product in 2026. The infrastructure underneath it was not. The orchestration layer, the audit trails, and the offline-first job queues keep feeding everything we build, and they are the foundation the current research programme stands on.

What the field changes

Field conditions rewrite the engineering. Nodes are ruggedised or they are replaced. Sensors report late, or not at all, and the system has to treat silence as normal rather than as an alarm. Power arrives in bursts, so work is queued and resumed instead of assumed continuous.

An agent that assumes it will always reach its model endpoint fails exactly when it is needed. The infrastructure we design now plans for the gap and recovers without a human in the loop.

Close view of a rusted steel shackle and bolt, salt-worn hardware that still holds — Corrosion is not failure. Hardware here is built to age in place and keep its load.

Where it goes from here

The line that NCRST boosted in the company's Gemsweb Digital era now runs in-house under the Tangison name. The direction has not changed: intelligence that runs where it is deployed, on hardware its owners control, under conditions nobody would call ideal.

Everything proven in the research lands in the systems we deliver. Applied AI that keeps running through outages. Data that never has to leave your walls. Automation you can audit line by line. That is not a feature list, it is the design constraint the whole company was built around.